SR Ground Dataset and Supplementary Material This repository accompanies the paper SR Ground: Image Quality Grounding for Super Resolved Content and provides the SR Ground dataset along with supplementary materials, including model training details, extended experiments, and annotation processes. Repository Structure datasets/ Contains all images. Each sample is located in a folder named according to the pattern: – name of the Super‑Resolution method used for upscaling. – scale factor applied to obtain the low‑resolution image. Each sample folder contains four files: Ground‑truth image – no suffix. Low‑resolution image – @LR@ suffix. Low‑resolution image upscaled with bicubic interpolation – @RF@ suffix. Super‑resolved image – @SR@ suffix. outputs/ Contains predictions from the Image Quality Grounding models. Each file follows the pattern: outputs/ / / .npy.gz , – same meaning as above. – distortion type segmented; one of real distortions or sr artifacts . – name of the super‑resolved image (matches the @SR@ file in datasets/ ). masks for markup.json A JSON dictionary that records which segmentation masks were refined through crowdsourcing. Keys: distortion types ( "real distortion…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy